GNN User Scheduling for Dynamic MU-MIMO Resource Allocation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional user scheduling methods for next-generation wireless networks, such as greedy algorithms and neural network solutions, face high complexity and performance issues due to dynamic user numbers and varying channel conditions in MU-MIMO systems.
Innovation Solution
Employing a graph neural network (GNN) to provide a scalable and comprehensive characterization of user interactions within the wireless network, allowing for efficient resource allocation through a one-shot user group selection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional user scheduling methods (greedy algorithms, neural network solutions) are used in MU-MIMO systems, then resource allocation can be performed, but the system complexity and processing time increase significantly due to dynamic user numbers and varying channel conditions
Solution Approach 1:
The patent transforms the user scheduling problem by changing the representation parameters from traditional sequential user lists to graph-structured data, where users are nodes and interference relationships are edges. This parameter transformation enables the GNN to process dynamic user sets efficiently without exponential complexity growth, directly resolving the contradiction between handling dynamic conditions and maintaining low computational complexity
Solution Approach 2:
The patent replaces conventional mechanical scheduling algorithms (greedy approaches, exhaustive search) with a neural network-based system that learns optimal scheduling policies. The GNN model substitutes iterative computational mechanics with parallel neural computation, achieving faster processing times while adapting to dynamic channel conditions and user movements
2Adaptability or versatility
If traditional neural network solutions are applied to user scheduling, then scheduling decisions can be made, but the complexity increases and performance degrades under dynamic user numbers and varying channel conditions
Solution Approach 1:
The patent designs a universal GNN-based scheduling framework that handles variable numbers of users through graph structure flexibility. The same model architecture processes different user set sizes by dynamically adjusting graph nodes, eliminating the need for multiple specialized models. This universal approach maintains adaptability while controlling complexity through efficient graph neural network operations
Solution Approach 2:
The patent transitions from one-dimensional sequential user processing to two-dimensional graph-structured representation, where users and their interference relationships form a spatial network. This dimensional transformation enables the model to capture complex user interactions efficiently, achieving high adaptability to dynamic conditions without proportional increases in computational complexity
3Reliability
If comprehensive user interaction characterization is achieved, then better weighted sum rates are obtained, but processing time increases
Solution Approach 1:
The patent implements continuous useful action through the GNN's ability to process user scheduling decisions in a streamlined manner. The graph neural network continuously updates user selections based on interference relationships without requiring repeated exhaustive evaluations, maintaining high weighted sum rate performance while reducing processing time through efficient iterative refinement of scheduling decisions
Data Source
AI summary
The present disclosure relates to a device for use in a wireless network, the device including: a processor configured to: provide input data to a trained graph neural network model, the input data being indicative of a graph representation of a plurality of wireless communication devices, wherein the trained graph neural network model is configured to provide output data being indicative of a scheduled user set including the plurality of wireless communication devices; and instruct user scheduling of the plurality of wireless communication devices based on the output data of the trained graph neural network model.